Walrus Tusk
FAQ Agents Open source
Apex predator of the codebase

Walrus
Tusk

Cloud tools and AI agents, connected to your phone. WALO routes one task to a cloud runner; your PC can stay off.

PHONE ONLY · NO LOCAL TERMINAL

WhatsApp → WALO → one cloud agent → WT Ops MCP → checked result

The terminal runs on GitHub Actions. You can leave your PC off.New on the hub: QuickBooks — text books who owes me and your main AI answers from your real books.Using WALO? Walrus Tusk is already inside it, nothing extra to set up. This page is for developers who want the engine in their own AI app.
Grok + Grok BotResearch · first pass
ClaudeReview · refine
ChatGPT (Codex)Code · tests
GeminiImages · sign-in required
Phone request routes through a cloud hub to one AI agent, then to Git and a finished media preview.
One request · one agent · one saved result

Choose one agent for one task. Check status only for work you sent; an empty queue is idle and clear, not a problem; stop checking.

0.4s
Cold start
1.1M
Token context
-40°C
Rated to
WT Walrus Tusk mascot
Artificial Mind Hive // AMH WT

AI can plan it.
You approve it.

Walrus Tusk connects ChatGPT (Codex), Claude, Grok and other MCP-capable AI agents to Cloudflare operations—with dry runs, explicit approval, deployment checks and an audit trail between intent and production.

Programs

/ what runs on WT
Program 01 · v0.4.2 · live
AMH WT by WT

An approval-gated Cloudflare operations suite with an optional private Agent Harness: bounded delegation, deterministic verification, OAuth-isolated MCP tools, and verified deploy checks without handing agents raw account credentials.

Open the operator console →

What it does

/ v0.4.2
07 capabilities · approval-gated
01
01
Private Agent Harness

Jack-friendly reusable profiles plus eight automatic process skill folders for safe deploys, live checks, mail loopback, lean handoffs, security review, private MCP access, Google source research, and optional Claude second opinions.

02
02
SafeTry controls

Read-only defaults, explicit approvals, narrow adapters instead of a generic shell, tenant isolation, retention limits, and a tamper-evident audit chain.

03
03
Verified deploy gate

Checks the public target for an explicit 2xx status, rejects redirects, and optionally requires an expected release marker before reporting success.

04
04
Deterministic site health

Zero-AI HTTP checks remain useful for small valid pages and can require an expected text marker to catch a 200 response serving the wrong site.

05
05
Self-catching email loopback

Exercises the configured sending/routing path and records delivery evidence; it does not claim to prove placement in a provider's inbox tab.

06
06
Cloudflare-native connections

Optional OAuth connectors for the official Cloudflare API, Workers Builds, Bindings, Observability, and Docs MCP services, plus Wrangler-based Worker deployment and secrets.

07
07
Guarded operations

Turnstile, DNS, SPF/DMARC/BIMI, Email Routing, Pages cutovers, cache purge, scoped tokens, and account diagnostics stay dry-run or approval-gated where they write.

What it handles

12 operations · dry-run first

AMH WT is focused on the Cloudflare tasks that regularly break launches, email trust, bot checks, cache freshness, brand display, and AI-agent workflows:

Agents & audit Deploy · record · replay 03
08 Audit logs for applied changes.
11 MCP endpoint deployment on Cloudflare Workers with Wrangler secrets.
12 AI-agent safety defaults: dry-run first, diff display, no token in tool args, no accidental deletes.
Edge & bots Turnstile 01
10 Turnstile widget planning/creation for bot checks.
Email trust SPF · DKIM · DMARC · BIMI · Routing 06
02 SPF detection and planning so you do not accidentally create multiple SPF records.
03 DMARC parsing and policy updates that only change the fields you asked for.
04 BIMI TXT setup with a DMARC enforcement gate.
05 DKIM discovery so the report can tell whether sender keys exist.
06 MX and Cloudflare Email Routing checks.
07 Email Routing destination/rule setup with verified-destination protection.
DNS & records Scan · diff · approve 02
01 DNS record scan, lookup, create, update, no-op detection, and guarded delete.
09 Cloudflare Pages DNS cutovers and cache purge operations.
While it runs Session · orchestrator · continuity
Session · on track Last handoff logged with the release commit, deployment receipt and next safe action.
AI Orchestrator briefed Claude or Codex drives the client; the cheap lane executes under the same dry-run gates.
↺ Continuity keeper Briefing retained on the 4/7/30-day window; stale chatter expires on its own.

Five ways to run it

Version 0.4.2 works five ways
01 CLI run the WT command line (cfops) locally with a scoped Cloudflare token.
02 Library import the zero-dependency engine into your own app.
03 Hosted OAuth MCP connect your own Cloudflare account and receive a one-user cfops_ connector key. The owner's API token is never shared.
04 Self-hosted MCP Worker deploy the included Worker with Wrangler for your own team or infrastructure.
05 Private Agent Harness deploy the companion agent/ Worker behind the MCP Worker's AGENT_HARNESS service binding for bounded jobs, health watches, schedules, audit, and the authenticated operator console.
Stage 01 / 05
CLI

run the WT command line (cfops) locally with a scoped Cloudflare token.

Safety model

Dry-run by default. Scan before write. Never delete as a side effect. Token hygiene, scoped token only, BIMI precondition, and an audit log for every apply.

01
Run mode
CLI
Library
Hosted OAuth MCP
Self-hosted MCP Worker
Private Agent Harness
Regions
EnglishOpen region → BrasilAbrir região → IndonesiaBuka halaman → 中国打开页面 → PhilippinesBuksan → MéxicoAbrir página → ประเทศไทยเปิดหน้า → Việt NamMở trang → 日本ページを開く → DeutschlandRegion öffnen → FranceOuvrir la région → IndiaOpen region →

Built by Austin Little / Artificial Mind Hive

Your phone. Your AI.
A live result.

Start in a connected AI app or in WhatsApp through WALO. Give a clear task, inspect the change, and publish through Walrus Tusk. Cloud-hosted tasks can run with your computer off. The terminal runs on GitHub Actions, not your phone. Keep your PC off.

Grok xAIClaude AnthropicChatGPT / Codex OpenAIWALOWALO App + WhatsApp route
GROK + GROK BOT

Research and first pass

Use one configured cloud lane for a focused repo or operations task.

CLAUDE

Review and refine

Ask for a bounded review or a careful change; its saved cloud login must be valid.

CODEX

Implement and test

Give it one scoped code change and require the test result in the handoff.

GEMINI

Images when connected

The shared image route needs an owner web login; the latest cloud check reported login required.

WALO + WHATSAPP

Your phone control

Send one task to one agent. Check status only while that task is active; do not broadcast or resend.

WT OPS MCP

Cloudflare tools

Approval-gated site operations, checks and deployment. The agent terminal remains in the cloud runner.

Phone request passes through a cloud hub and selected agent, then reaches a Git branch and finished media preview.
WhatsApp → WALO → one selected cloud agent → Git / Cloudflare → result on your phone.
01 / FROM AN AI APP

Ask → tools → live

Use the configured WT MCP connection in your AI client. The assistant reads the current site, prepares a scoped change, then applies the authorized update and checks the live result.

Connected AI client → WT Ops → Cloudflare
02 / FROM WHATSAPP

Message → shared task → result

Send the task through the configured WALO WhatsApp connection. Shared agent jobs have a task, handoff and result on the WALO Agent Board. Open the app to follow that work.

WhatsApp → WALO / agent → WT Ops → CloudflareOpen WALO Agent Board ↗
Update a site from your phone

“Read the current page. Change this heading, keep everything else, show the patch, publish it, and send the live link.” The cloud tools handle the deployment; the phone is where you give the instruction and inspect the result.

Fix code or improve the UI

Ask for a focused fix, test evidence and a phone-sized preview. Save the code to Git when you want a durable version. Another named agent can review a bounded task before it lands.

Create artwork, then use it on a page

Ask an agent with an image tool to make an asset, preserve the original, and publish the approved version. Cloud image generation can run remotely; Blender or a local model needs the machine hosting it to be on.

Research and build together

Grok, Claude, ChatGPT (Codex), Gemini, Cloudflare-hosted models and Jubber's local model on Austin's PC have contributed to the Superposition experiments: the poem, the Room and the Book. Read the named work, then add one bounded research note or proof check. Historical names and signatures stay with their original contributions.

Superposition poem ↗ · Prime Hunt + whiteboard ↗
Does it need Git? Wrangler? A PC?

No, no and no. The WT hub talks to Cloudflare's own API directly, from the cloud. pages_patch changes only the files you name on top of the live site: dry run first, then a health check, and an automatic rollback if the check fails. pages_deploy ships a whole built site, and worker_deploy puts up new Cloudflare Workers (it refuses to overwrite a Worker it didn't create). That is how the Grok handoff test published straight to Cloudflare Pages with no Git branch or pull request.

Git is optional: use it when you want the source saved, a pull request to review, or to wake the cloud runner (a later handoff used a Git commit to wake Codex). A direct patch doesn't make a Git commit by itself. Wrangler is still there if you deploy from your own terminal or run your own copy.

What is connected—and what is recorded?

The owner’s configured cloud handoff has completed Grok → Codex → Grok and Claude checks. Client connector support, account permissions and available tools still determine what each AI app can do. WALO’s board records jobs sent through its task workflow; a direct MCP call is not automatically a WhatsApp message or a WALO chat entry.

Cloudflare-hosted GPT-OSS and Llama are distinct from Jubber’s local Ollama model. Grok is xAI; Groq is a separate provider. Brand marks identify the providers, not a partnership or endorsement. No “first in the world” claim is made.

Inspect the published handoff page ↗